Privacy and imagery terms
Short version: the location you paste is used to fetch imagery, then discarded. It is never stored, never written to a log, and never sent to the model.
What happens to the location you enter
The coordinates are held in memory for the lifetime of one analysis and used for exactly two things: fetching imagery, and — after the prediction already exists — looking up which country the point falls in so you can compare the two.
An analysis expires ten minutes after it starts, at which point the coordinates and the rendered views are dropped. Nothing is written to a database or to disk; there is no persistence layer in this application at all.
What the model receives
Images, a flag saying which of them are usable, and the camera angles they were taken at. That is the entire payload, and the request schema has no field capable of holding anything else — a request carrying a coordinate, a panorama id or a URL is rejected rather than quietly ignored.
Image metadata is stripped before preprocessing, so EXIF GPS tags cannot reach the model even when a source image carries them.
Logging
Logs record a request identifier, which stage ran, and how long it took. They do not contain coordinates, the URL you pasted, panorama identifiers or image data. No analytics or error-reporting service receives your input.
Google Maps links
A Google Maps or Street View URL is accepted as a way of expressing coordinates you already have. The link is parsed as text, offline. Nothing fetches google.com/maps, its tile endpoints, its panorama endpoints, or any thumbnail.
Google Street View imagery is not used as a model input and is not used for training, testing or benchmarking. That path exists in the code behind a disabled flag so the boundary is explicit and testable; enabling it would require written authorisation naming that specific use.
Imagery in this deployment
The default provider serves synthetic panoramas generated by this repository. They depict invented countries, contain no real imagery and no real people, and are freely redistributable.
Every provider declares what may be done with its imagery — inference, training, caching, attribution — and those declarations are enforced in code rather than only documented. The per-provider verdicts, with the terms reviewed and the date, are in docs/imagery-compliance.md.
What a prediction is not
A prediction is a probabilistic statement about what a place looks like. It is not evidence of any person’s location, nationality or movements, and it should not be used to make decisions about anyone.